BibTeX
@MISC{Hough_facultyof,
author = {Julian Hough and Matthew Purver},
title = {Faculty of Linguistics and Literature},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We present STIR (STrongly Incremen-tal Repair detection), a system that de-tects speech repairs and edit terms on transcripts incrementally with minimal la-tency. STIR uses information-theoretic measures from n-gram models as its principal decision features in a pipeline of classifiers detecting the characteristic properties of different stages of repairs. Results on the Switchboard disfluency tagged corpus show utterance-final accu-racy on a par with state-of-the-art incre-mental repair detection methods, but with better incremental accuracy, faster time-to-detection and less computational over-head. We evaluate its performance using incremental metrics and propose new re-pair processing evaluation standards. 1
Keyphrases
state-of-the-art incre-mental repair detection method n-gram model edit term corpus show utterance-final accu-racy incremental metric incremen-tal repair detection switchboard disfluency different stage computational over-head information-theoretic measure new re-pair processing evaluation standard speech repair minimal la-tency incremental accuracy principal decision feature characteristic property
